In complex workflows, tasks often need to be executed in a specific order. Some tasks may rely on the outputs of others, or there might be a logical sequence that must be followed to achieve the desired outcome. ControlFlow provides several mechanisms to define and manage these task relationships, ensuring that your workflow executes in the correct order and that data flows properly between tasks.

ControlFlow offers two primary ways to establish relationships between tasks: sequential dependencies and subtask relationships. Each method has its own use cases and benefits, allowing you to structure your workflows in the most appropriate way for your specific needs.

​
Upstream dependencies

Upstream dependencies are the most straightforward way to specify that one task must wait for another to complete before it can begin. This is done using the depends_on parameter when creating a task.

import controlflow as cf

@cf.flow
def research_flow():
    gather_sources = cf.Task("Gather research sources", result_type=list[str])
    
    analyze_sources = cf.Task(
        "Analyze gathered sources",
        result_type=dict,
        depends_on=[gather_sources]  # explicit dependency
    )
    
    return analyze_sources

result = research_flow()
print(result)

In this example, analyze_sources will not start until gather_sources has completed successfully.

​
Subtasks

Subtasks create a hierarchical dependency structure. A parent task can not be completed until all of its subtasks have finished. This hierarchical structure enables you to create detailed, step-by-step workflows that an AI agent can follow, ensuring thorough and accurate task completion.

​
Imperative creation

You can create subtasks imperatively by passing the parent task as an argument when creating a new task:

import controlflow as cf

parent_task = cf.Task("Create a greeting")

t1 = cf.Task("Choose a greeting word", parent=parent_task)
t2 = cf.Task("Add a friendly adjective", parent=parent_task, depends_on=[t1])
t3 = cf.Task("Construct the final greeting", parent=parent_task, depends_on=[t2])

result = parent_task.run()
print(result)

​
Context managers

Another way to create subtasks is by using a context manager. This approach allows you to dynamically generate and execute subtasks within the scope of a parent task.

import controlflow as cf

with cf.Task("Create a greeting") as parent_task:
    t1 = cf.Task("Choose a greeting word")
    t2 = cf.Task("Add a friendly adjective", depends_on=[t1])
    t3 = cf.Task("Construct the final greeting", depends_on=[t2])

result = parent_task.run()
print(result)

​
Automatic Execution of Dependencies

A key feature of ControlFlow’s dependency management is that you don’t need to explicitly run dependent tasks. When you run a task, ControlFlow automatically executes all of its dependencies, including:

  • Tasks specified in the depends_on parameter
  • Subtasks (for parent tasks)

This means that when you run a flow or task, you only need to run or return the final task(s) in the workflow DAG. ControlFlow will ensure that all necessary upstream tasks and subtasks are executed in the correct order.

For example:

import controlflow as cf

@cf.flow
def research_flow():
    gather_sources = cf.Task("Gather research sources", result_type=list[str])
    
    analyze_sources = cf.Task(
        "Analyze gathered sources",
        result_type=dict,
        depends_on=[gather_sources]
    )
    
    write_report = cf.Task(
        "Write research report",
        result_type=str,
        depends_on=[analyze_sources]
    )
    
    # Only need to run the final task
    return write_report.run()

research_flow()

In this example, running write_report will automatically trigger the execution of analyze_sources, which in turn will trigger gather_sources. You don’t need to explicitly run or return gather_sources or analyze_sources.

To learn more, see running tasks.